The secrets of Effective Prognostics

08 November 2017

It’s a fact that no-matter how advanced the machine, there will come a time when it unexpectedly fails and your factory experiences the financial and reputational losses caused by unplanned machine downtime. The impact of this can vary dramatically; from direct costs of over £1.5M per hour in automotive manufacturing to thousands per hour in the Fast-Moving Consumer Goods (FMCG) industry, not to mention damage to reputation. Even machines cared for under the best preventative maintenance regime will experience unplanned failure at some point in their lifetime. 

Condition monitoring – the answer?

Condition Monitoring(CM) is the process of monitoring time-series data (vibration, acoustic emissions, temperature, etc.) related to machinery in order to analyse it and identify changes which may indicate faults. 

Condition monitoring project life cycles tend to follow a bespoke (expensive) engineering roadmap from specifying the system through implementation, rollout, training and ongoing support. It is not until the later stages of this roadmap that value is achieved and a Return on Investment (ROI) can be demonstrated, if at all. Condition monitoring unfortunately also requires expert manual analysis and critically the data itself has no context – it doesn’t help maintainers or anyone else to understand for how long the machinery will continue to work. 

The limited scalability of condition monitoring and often unknowable ROI makes it difficult to convince budget-holders to invest. Prognostics however provides the crucial missing information.

Prognostics allows you to understand Remaining Useful Life
Prognostics allows you to understand the Remaining Useful Life of machinery

Prognostics is the science of accurately forecasting the Remaining Useful Life (RUL) of machinery. This added contextual information makes it useful for everyone involved in plant operations, from the maintainer to the CFO and it’s the key to enabling accurate predictive maintenance. However, it’s difficult to get right and there are a number of key points to carefully consider: 

It must be driven by condition monitoring

Prognostics needs condition monitoring data in order to work. That means that an investment in the basic hardware, communications and data storage infrastructure to support condition monitoring does need to be in place. These should be selected with scalability (ease of installation, minimal maintenance, minimal cost) in mind. It may be possible to extract data from machine PLCs, avoiding the need for additional hardware investment during an initial project. 

It needs accurate degradation information

This can be captured in a number of ways but till recently the most popular method has been to run assets to destruction and record their failure modes for incorporation into hard-coded models. Not just for cost reasons this is obviously impractical in an industrial scenario, as such this technique has mostly been limited to the aerospace & defence industries with obvious safety-cases.

 A far more practical approach is to use maintenance information to build prognostic models. This maintenance information must be stored electronically in a maintenance management system if automated prognostics is to be used. 

To be effective, it must be automated

Like condition monitoring, prognostics driven by human calculation and interpretation cannot scale. The skills required typically involve a deep understanding of condition monitoring and data analysis techniques which usually come with a heavy price tag (£50,000+). This may be fine for less than 50 assets but manual analysis quickly becomes unreasonably expensive if tied to human resources. Industries such as Aerospace & Defence can afford this overhead however it’s beyond most industrial companies to have a team of dedicated data scientists per group of assets.

Thankfully, the recent explosion of inexpensive cloud computing and availability of machine learning techniques as made available the kind of computational power required to automate prognostics.

Use large number of similar assets

The manufacturing sector typically has a large number of highly utilised machines which tend to be the same or similar across geographies and sub-sectors.  This gives greater statistical significance in the analysis of the data, as well as the ability to share prognostic models and analysis techniques. 

Having a proven failure identification method for a machine provides benefit for that machine as well as to all similar machines.  This provides economies of scale, dramatically reducing the costs of prognostic systems.

Contextualise with events

In addition to sensor data, smart factories are rich in data related to utilisation, maintenance and other operational events that can give additional context which is helpful for prognostic models.  Understanding the changes to machine state due to production demands aids in identifying the difference between an emergent failure and simply a change of operation.

Implementing prognostics

With so many IoT related projects failing, it’s important to plan correctly to avoid your prognostics project adding to an unfortunate statistic. Prognostics is challenging but to increase the likelihood of success, we suggest following the ‘3 Es’ approach: Establish, Exploit and Enhance:

establish exploit enhance

Establish

Extracting and utilising the data and sensing capability already in PLCs in the Establish phase allows a prognostics project to start with minimal investment, with the objective of establishing a capability that can be exploited to deliver a positive result and benefit to the business. This should be performed on a small set of machines with clear failure issues and a Mean Time Between Failure of 6 months or less so that failure can be observed and learned from during the next phase.

Exploit

During the Exploit phase and initially on a reduced scale, evidence is gathered to define the gaps in the prognostic solution along with evidence of downtime reduction and added machine lifetime context.  This evidence drives the definition of the Enhance phase.

Enhance

With the greater understanding of a prognostics problem and the ways in which it provides benefit, the Enhance phase will have the required evidence to justify spend on additional sensing hardware and capability for the next iteration through the ‘3 Es’ cycle. It’s important here to specify a scalable prognostics software product.

Once the Enhancements are implemented, the cycle can be repeated to increase the scale of the capability and enjoy a greater ROI.

The benefits brought by prognostics

In a study conducted recently by McKinsey, it was revealed that a predictive maintenance program could expect to reduce downtime by over 50%.

For a company operating at a respectable Overall Equipment Effectiveness (OEE) of 50% (85% Quality, 69% Availability, 85% Performance), using automated prognostics to drive predictive maintenance could quickly improve OEE by a huge 11%, before taking into account the indirect improvements that will be seen in the Quality and Performance metrics. 

Effective prognostics allows organisations to dramatically reduce unplanned downtime; decreasing costs and improving profitability.

 Iterative prognostics does not require a vast capital budget needing high level approval and has a short lead time where benefits can be rapidly demonstrated in order to gain approval for scaling. The key advice is to avoid the temptation of the ‘big bang’; start in a focused way, demonstrate success and use that as a justification for scaling.

Prognostics is a revolution for the usability of condition monitoring but it needs to be introduced through evolution.

Want to learn more about prognostics and the benefits it can bring to your organisation? Download your FREE prognostics white paper:

Free prognostics white paper download

what is prognostics

Keep me notified

Most Read